Generative AI and ChatGPT
AA: What does ChatGPT and generative AI mean for AI and data?
JTW: The architecture of transformers is 1) is adaptable to multiple purposes, and 2) can be multi-modal. This marks a drastic move away from the era of single purpose models and early steps away from what everyday language calls “narrow” AI. This also unlocks multi-modal uses cases and I expect a sharp increase in applications that incorporate and integrate multiple data types (for example, text and images)
The way that running generative models like ChatGPT multiple times with the same input yields different results, together with the multi-purpose nature of such models, means that governance and testing will have to undergo a paradigm shift. Governance on the underlying models of ChatGPT is quite meaningless as it can be used to generate both pizza or poison. Governance will then have to shift to the point where the model is fully formed into a use case and /or to the point...